Expertise Networks in Online Communities: Structure and Algorithms

Save this PDF as:

Size: px
Start display at page:

Download "Expertise Networks in Online Communities: Structure and Algorithms"


1 WWW / Tck: E*-Applictions Sssion: E-Communitis Exptis Ntwoks in Onlin Communitis: Stuctu nd Aloithms Jun Zhn School of Infomtion Univsity of Michin Mk S. Ackmn Dpt. of EECS nd School of Infomtion Univsity of Michin Ld Admic School of Infomtion Univsity of Michin ABSTRACT Wb-bsd communitis hv bcom impotnt plcs fo popl to sk nd sh xptis. W find tht ntwoks in ths communitis typiclly diff in thi topoloy fom oth onlin ntwoks such s th Wold Wid Wb. Systms ttd to umnt wb-bsd communitis by utomticlly idntifyin uss with xptis, fo xmpl, nd to dpt to th undlyin intction dynmics. In this study, w nlyz th Jv Foum, l onlin hlp-skin community, usin socil ntwok nlysis mthods. W tst st of ntwok-bsd nkin loithms, includin PRnk nd HITS, on this l siz socil ntwok in od to idntify uss with hih xptis. W thn us simultions to idntify smll numb of simpl simultion uls ovnin th qustion-nsw dynmic in th ntwok. Ths simpl uls not only plict th stuctul chctistics nd loithm pfomnc on th mpiiclly obsvd Jv Foum, but lso llow us to vlut how oth loithms my pfom in communitis with diffnt chctistics. W bliv this ppoch will b fuitful fo pcticl loithm dsin nd implmnttion fo onlin xptis-shin communitis. Ctois nd Subjct Dsciptos H... [Infomtion Intfcs nd Psnttion (.., HCI)]: Goup nd Oniztionl Intfcs collbotiv computin, comput-suppotd cooptiv wok, thoy nd modls, wbbsd intction. J. [Comput Applictions] Gnl. Gnl Tms Humn Fctos, Aloithms, Expimnttion. Kywods Socil ntwok nlysis, xptis findin, xpt loctos, hlp skin, onlin communitis, simultion 1. INTRODUCTION Stv is Jv pomm who just sttd wokin on pojct usin Jv Spch on nw mobil pltfom. But h cnnot un his fist Jv Spch pom on th nw pltfom nd nds som hlp. Stv is unbl to tll whth th poblm hs isn bcus h dos not undstnd how to us th Jv Spch pck, o bcus Jv Spch dos not suppot th mobil pltfom wll. Copyiht is hld by th Intntionl Wold Wid Wb Confnc Committ (IWC). Distibution of ths pps is limitd to clssoom us, nd psonl us by oths. WWW, My 1,, Bnff, Albt, Cnd. ACM -1---//. It cn b difficult to t stisfctoy nsw to Stv s poblm by schin Gool dictly. Instd, h my pf to find nd sk somon who hs ltd xptis o xpinc, nd onlin communitis hv md s on of th most impotnt plcs fo popl to sk dvic o hlp. Th topics n fom dvic on mdicl ttmnt, pommin, softw, buildin comput fom sctch to piin th kitchn sink. Ths communitis usully bound by shd pofssions, intsts, o poducts mon thi pticipnts. Fo instnc, th Sun Jv Foum hs thousnds of Jv dvlops comin to th sit to sk nd nsw qustions ltd to Jv pommin vy dy. Th Micosoft TchNt nwsoup is mjo plc fo pomms to sk hlp fo pommin qustions ltin to Micosoft poducts. Evn thouh uss in ths onlin communitis usully do not know ch oth nd idntifid usin psudonyms, thy willin to hlp ch oth fo vious sons, such s ltuism, puttion-nhncmnt bnfits, xpctd cipocity, nd dict lnin bnfits [1, 1]. This wok sks to nhnc onlin communitis with xptis finds. Exptis finds, o xptis loction nins, systms tht hlp find oths with th ppopit xptis to nsw qustion. Ths systms hv bn xplod in sis of studis, includin Stt nd Lochbum [], Kulwich nd Buky [], nd McDonld nd Ackmn [] s wll s th studis in Ackmn t l. []. Nw systms, which us socil ntwok to hlp find popl, hv lso bn xplod, most notbly in Ynt [1], RflWb [1], nd most cntly commcil systms fom Tcit nd Micosoft. Ths systms ttmpt to lv th socil ntwok within n oniztion o community to hlp find th ppopit oths. Asid fom lyin on socil ntwoks, noth intstin chctistic of ths systms is tht thy tnd to blu th dichotomy btwn xpts nd sks. Thy tt on s xptis s ltiv concpt []. In lity, ltivly fw popl will clim thmslvs s n xpt, but mny popl tht thy hv som msu of xptis in som. Ths systms llow vyon to contibut s thy cn. Fo ths xptis find systms to b of sinificnt ssistnc, thy must ffctivly idntify popl who hv xptis in th dsid by th sk. Most cunt systms us modn infomtion tivl tchniqus to discov xptis fom implicit o scondy lctonic soucs. A pson s xptis is usully dscibd s tm vcto nd is usd lt fo mtchin xptis quis usin stndd IR tchniqus. Th sult usully is list of ltd popl with no intinsic nkin od o nks divd fom tm fquncis. It my flct whth pson knows bout topic, but it is difficult to distinuish tht pson s ltiv xptis lvls. Rlyin on wod nd documnt fquncis hs povn to b limitd []. 1

2 WWW / Tck: E*-Applictions Sssion: E-Communitis To mliot this, Cmpbll t l. [] nd Dom t l. [] usd ph-bsd nkin loithms in ddition to contnt nlysis to nk uss xptis lvls. This wok, don t IBM Rsch, pplid svl ph-bsd loithms, includin PRnk nd HITS, to both synthtic ntwok st nd smll mil ntwok to nk cospondnts ccodin to thi d of xptis on subjcts of intst. Thy found tht usin ph-bsd loithm ffctivly xtcts mo infomtion thn is found in contnt lon. Howv, th is wknss in ths studis. Th siz of thi ntwoks is vy smll nd dos not flct th chctistics of listic socil ntwoks. As sult, w wishd to visit th possibilitis of usin phbsd loithms on socil ntwoks of uss in onlin communitis. In this study, w nlyz l onlin hlp skin community, th Jv Foum, usin socil ntwok nlysis mthods. W thn tst st of ntwok-bsd loithms, includin PRnk nd HITS, on this l siz socil ntwok. Usin st of simultions, w xplo how vious ntwok stuctus ffct th pfomnc of ths loithms. W find smll numb of stuctul chctistics in th socil ntwoks tht w bliv ld to diffncs in th loithms' pfomnc fo onlin communitis. W xpct tht not only will ths chctistics b fuitful fo pcticl loithm dsin nd implmnttion, but tht thy will off nw sch insihts fo oths to xplo. Th pp pocds s follows. In Sction, w intoduc th community xptis ntwok nd bifly viw ltd wok. In sction, w dscib th ntwok chctistics of ou tst onlin community, th Jv Foum. In sction, w dscib som xptis nkin loithms. In sction, w psnt n vlution compin th nkins poducd by humn ts nd by th loithms. In sction, w thn xplo th ntwok chctistics tht ffct th pfomnc of ths loithms usin simultion study. And finlly, w summiz ou findins in Sction.. EXPERTISE NETWORK IN ONLINE COMMUNITIES Onlin communitis usully hv discussion thd stuctu. A us posts topic o qustion, nd thn som oth uss post plis to ith pticipt in th discussion o to nsw qustion posd in th oiinl post. Usin ths postin/plyin thds in community, w cn ct post-ply ntwok by viwin ch pticiptin us s nod, nd linkin th ID of us sttin topic thd to pli s ID, s shown in Fiu 1. Uss Topic Fiu 1: W mp plyin ltionship into dictd ph. On th lft w hv biptit ph of uss (cicls) nd th discussion thds (squs) thy pticiptd in. This is tnsfomd to dictd ph wh n d is dwn fom th us mkin th initil post (th dshd d shown in n) to vyon who plid to it. This post-ply ntwok hs som intstin chctistics. Fist, it is not intntionlly built by its uss fo th pupos of fomin tis. Thus, it is not ntwok focusd on socil ltionships. Instd, it flcts community mmbs shd intsts. Whth it is community cntd on qustions nd nsws, socil suppot, o discussion, th son tht us plis to topic is usully bcus of n intst in th contnt of th topic th thn who sttd th thd. This indictly flcts pticul shd intst btwn th oiinl post nd th plis (lthouh th plis sntimnt bout th topic my diff). Futhmo, in qustion nd nsw community, th diction of th links cis mo infomtion thn just shd intst. A us plyin to noth us s qustion usully indicts tht th pli hs supio xptis on th subjct thn th sk. Th distibution of xptis, lon with th ntwok of sponss, is wht w will cll th community xptis ntwok (CEN). It indicts wht xptis xists within n onlin community, s wll s how it is distibutd in pctic. Th full dynmic of CEN my b much complx in som communitis. Fo xmpl, th my b tolls, spmms, tc. An nsw thd to qustion cn b th sult of complx socil pocss nd th fist fw plis my ctully not nsw th qustion but ty to clify th poblm. Th ntwok could b wihtd ccodin to th fquncy of how oftn us hlps noth. W will discuss ths issus in lt sctions. Stuctul Psti in Socil Ntwoks Exptis is closly ltd to stuctul psti msus nd nkins in socil ntwok studis. In dictd ntwoks, popl who civ mny positiv choics considd to b pstiious, nd psti bcoms slint spcilly if positiv choics not cipoctd []. Rschs in vious filds hv pplid ths psti ids to diffnt typs of ntwoks. Fish t l. [11] usd socil ntwok visuliztion nd nlysis on th pttns of plis fo ch utho in slctd nwsoups to find diffnt typs of pticipnts. Fo instnc, thy usd th ind (how mny popl us plid to) nd outd (how mny popl plid to th us) of us s ocntic ntwok to idntify th ols within th oup (.., nl sk o pli). Bolln t l. [] usd simil nkin msu to vlut th psti of cdmic jounls. Liu t l. [] usd it to vlut th impct of n individul utho in co-uthoship ntwok. And, of cous P t l. [] usd PRnk to nk wb ps. In onlin hlp-skin communitis, th socil ntwok is n xptis ntwok. Bcus th wy links constuctd, th psti msu of th ntwok is hihly coltd with us s xptis. Thus, this hints tht th oppotunitis to mk us of such ntwok stuctus to nk popl s xptis in onlin communitis, nd build ltd pplictions/systms tht futh impov th xptis shin in th onlin wold. Nxt w tun to th invstition of n xptis ntwok in on onlin community, th Jv Foum.. EMPIRICAL STUDY OF AN ONLINE COMMUNITY.1 Th Jv Foum Th Jv Dvlop Foum is n onlin community wh popl com to sk qustions bout Jv. It hs sub-foums tht focus

3 WWW / Tck: E*-Applictions Sssion: E-Communitis on vious topics concnin Jv pommin. Th is l divsity of uss, nin fom studnts lnin Jv to th top Jv xpts. Uss usully cn t n nsw ltivly quickly bcus of th l numb of pticipnts. In this study, w usd th Jv pommin sub-foum (clld h "Jv Foum"), which is plc fo popl to sk nl Jv pommin qustions. Th Jv Foum hd totl of,1 msss in, thds. W usd th ntwok constuctd upon ths thds to vlut th usfulnss of ou xptis-nkin loithms. Th Jv Foum ntwok hd 1, nods nd,1 ds. Th nxt sction dscibs th chctistics of th Jv Foum ntwok. This will povid both tst bd fo th loithms nd, lt in th pp, will hlp in undstndin th undlyin ntwok chctistics tht xptis nkin loithms opt upon.. Chctizin th Ntwok..1 Th Bow ti stuctu nlysis Not ll uss in th Jv Foum sk qustions, no do ll uss nsw qustions. Usin bow ti stuctu nlysis, w xmin th nl stuctu of th Jv Foum ntwok. Th bow ti stuctu, fist poposd by schs t IBM, AltVist, nd Compq, yilds insihts into th complx oniztion of th Wb ntwok stuctu. Its ky id is tht th wb is bow ti nd hs fou distinct componnts: Co, In, Out, nd Tndils nd Tubs (s Bod t l. []). In ou bow ti modl, cntl co contins uss tht fquntly hlp ch oth. It is stonly connctd componnt (SCC), mnin tht on cn ch vy us fom vy oth by followin qustion-nsw links. Th 'In componnt contins uss tht usully only sk qustions. Th Out consists of uss tht usully only nsw qustions postd by uss in th Co. Oth uss, th 'Tndils' nd 'Tubs', connct to ith th 'In o Out clusts, o both, but not to th Co. Thy uss who only nsw qustions posd by 'In uss o whos qustions only nswd by Out uss. Fiu, nd Tbl 1 comp th bow ti stuctu of th Jv Foum ntwok with tht of th Wb (s potd in []). Fiu : Th wb is bow ti Fiu : Th Jv Foum ntwok is n unvn bow ti Tbl 1: Compison of bow ti nlysis btwn Wb nd th Jv Foum ntwok Co In Out Tndils Tubs Disconnct Wb.% 1.% 1.% 1.%.%.% Foum 1.%.% 1.%.%.% 1.% Ths sults show th Jv Foum ntwok looks much diffnt fom th Wb. Th Jv Foum hs much bi In componnt nd ltivly smll Co thn th Wb. This indicts tht in this onlin community, only bout 1% of uss ctivly sk nd nsw qustions fo ch oth. Mo thn hlf of th uss usully only sk qustions, nd bout 1% uss usully only nsw qustions. This sult lso indicts tht instd of bin public plc wh popl hlp ch oth cipoclly, this onlin hlp skin community is mo closly plc wh sks com to sk hlp fom volunt hlps... Distibution of d W cn us th bow ti stuctu to show th ol of uss in th ntwok, but it dos not cptu th lvl of thi intction. Lookin t d distibutions is nl wy to dscib uss ltiv connctdnss in l complx ntwok [1]. Th d distibution is function dscibin th numb of uss in th ntwok with ivn d (numb of nihbos). An intstin common ftu of mny known complx ntwoks is thi scl-f ntu. In scl-f ntwok, th mjoity of nods ch connctd to just hndful of nihbos, but th fw hub nods tht hv dispopotiontly l numb of nihbos. Fiu shows th ind distibution histom fo th Jv Foum ntwok. It is hihly skwd (nd in fct scl-f xcpt fo cutoff t vy hih ds), simil to distibution obsvd fo Wb ps nd fo co-uthoship ntwoks. Th scl-f d distibution is flction of th hihly unvn distibution of pticiption. Instd of vybody hlpin ch oth qully, in th Jv Foum, th som xtmly ctiv uss who nsw lot of qustions whil mjoity of uss nsw only fw. Likwis, mny uss sk only sinl qustion, but som sk dozn o mo. b b o p v i t l u m u c = 1. fit, R =. ind outd 1 d of vtx Fiu : D distibution of th Jv Foum ntwok.. D coltions Whil th ind distibution shows how mny popl ivn us hlps, it ivs no infomtion bout thos uss' own tndncy to povid hlp. Fo xmpl, on miht lik to know whth hih volum plis only ply to nwbis, o if thy mostly tlk to oths simil to thmslvs. W cn nsw both of ths qustions by lookin t th coltion pofil (s Mslov t l. []) H w consid simplifid coltion pofil tht fo ch sk-pli pi counts th ind of th pli vsus th ind of th sk, s shown in Fiu. W lso pot simpl coltion cofficint btwn th sks' nd hlps' ind. Positiv ssottivity is common in socil ntwoks, wh popl with mny connctions tnd to know oth popl with mny

4 WWW / Tck: E*-Applictions Sssion: E-Communitis connctions whil hmits tnd to know oth hmits. W find howv, tht th Jv Foum is f fom n xclusiv club wh hih volum plis cospond with oth hih volum plis, lvin th nwbis to tlk to on noth. Rth, th Jv foum is nith ssottiv no disssottiv. Th coltion cofficint is v so slihtly ntiv t -.1, nd th coltion plot shows tht th hihst d nods (usully th xpts) tnd to nsw qustions coss th bod fom whov sks thm. As on miht xpct, low d uss (ons who pobbly lck th xptis to nsw oths qustions) typiclly do not ply to hih-d uss. o 1 l ( d n i k s hlp ind (loithmiclly binnd) Fiu : Th coltion pofil of th Jv Foum ntwok. Th colo cosponds to th loithm of th fquncy of such d piins. In summy, fom ths ntwok nlyss, w cn s tht th Jv Foum ntwok hs som uniqu chctistics, includin: Diffnt oups of uss fll into stuctully distinct pts of th ntwok: Th is bi 'In oup nd ltivly smll Co nd 'Out oups. Th uss ind distibution is skwd, with fw uss nswin l numb of qustions whil th mjoity of uss only nsw fw. Top plis nsw qustions fo vyon. Howv, lss xpt uss tnd to nsw qustions of oths with low xptis lvl. Sinc ths chctistics diffnt fom th Wold Wid Wb ph, thy cn potntilly ffct th pfomnc of vious xptis nkin loithms, s w will discuss nxt.. EXPERTISE RANKING ALGORITHMS Aft constuctin n xptis ntwok fom th post-ply pttns in th onlin community, nd hvin discovd intstin ulitis in th stuctu of th ntwok which miht colt with us s xptis, w now psnt svl loithms dsind to utomticlly inf us s xptis lvl. Aft psntin th loithms, w will povid th sults of thi tsts..1 Simpl Sttisticl Msus W sumis tht if pson nsws lot of qustions on topic, it is oftn th cs tht h o sh knows th topic wll. Excptions includ spmms who my b postin dvtismnts o tolls 1 who my b mkin inflmmtoy o othwis disuptiv posts. W found littl tollin o spmmin bhvio on th Jv Foum. Howv, ou obsvtions h would lso b pplicbl to foums wh spmmin is mo pvlnt, but cn b cubd o idntifid thouh uss lvnc fdbck. Rtunin to th Jv Foum, th simplst mthod fo vlutin us s xptis my b countin th numb of qustions nswd. W cll it th AnswNum msu. A slihtly diffnt msu is countin how mny oth uss us hlpd. Som uss my hv bi AnswNum but ll ths plis nswin qustions ptdly fom svl spcific uss. On th oth hnd, us who posts fw nsws, but in th pocss hlps t numb of uss, could hv bod o t xptis. Thus, countin how mny popl on hlps my b btt indicto thn countin th numb of plis. In socil ntwok, this could b clcultd usin th ind of nod.. Z-sco Msus Whil plyin to mny qustions implis tht on hs hih xptis, skin lot of qustions is usully n indicto tht on lcks xptis on som topics. Thus, w popos th z-sco s msu tht combins on s skin nd plyin pttns, s shown in followin fomul: If us mks n=q+ posts, q of thm qustions nd of thm nsws, w would lik to msu how diffnt this bhvio is fom ndom us who posts nsws with pobbility p =. nd posts nw qustions with pobbility 1-p =.. W would xpct such ndom us to post n*p = n/ plis with stndd dvition of n * p * (1 p) = n /. Th z-sco msus how mny stndd dvitions bov o blow th xpctd ndom vlu us lis: n / q z = = n / + q If us sks nd nsws bout qully oftn, thi z-sco will b clos to. If thy nsw mo thn sk, th z sco will b positiv, othwis, ntiv. W clcult th z-sco fo both th numb of qustions on skd nd nswd nd th numb of uss on plid to nd civd plis fom, dnotd sptly s Z_numb nd Z_d.. ExptisRnk Aloithm Th is potntil poblm in countin th numb of nsws on postd o th numb of popl on hlpd. A us who nsws nwbis qustions will b nkd s qully xpt s noth us who nsws dvncd uss qustions. Obviously th ltt usully hs t xptis thn th fom. Th wll known PRnk loithm, poposd by P t l. [] fo nkin wb ps, impovs this. It povids kind of p ssssmnt of th vlu of Wb p by tkin into ccount not just th numb of ps linkin to it, but lso th numb of ps pointin to thos ps, nd so on. Thus, link fom popul p is ivn hih wihtin thn on fom n unpopul p. Intuitivly, th nkin in PRnk cosponds to th fction of tim ndom wlk would spnd visitin p by ittivly followin links fom p to p. Th vious vsions of PRnk o simil msus; fo n ovviw, s [, ]. W popos usin PRnk-lik loithm to nt msu tht not only consids how mny oth popl on

5 WWW / Tck: E*-Applictions Sssion: E-Communitis hlpd, but lso whom h/sh hlpd. W cll it ExptisRnk. Th intuition bhind ExptisRnk is tht if B is bl to nsw A s qustion, nd C is bl to nsw B s qustion, C s xptis nk should b boostd not just bcus thy w bl to nsw qustion, but bcus thy w bl to nsw qustion of somon who hslf hd som xptis. In sns, ExptisRnk popts xptis scos thouh th qustionnsw ntwok. Tbl lists th ExptisRnk loithm tht is simil to PRnk. Tbl : Bsic ExptisRnk loithm Assum Us A hs nswd qustions fo uss U 1 U n., thn th ExptisRnk (ER) of Us A is ivn s follows: ER(A) = (1-d) + d (ER(U 1 )/C(U 1 ) + + ER(U n )/C(U n )) C(U i ) is dfind s th totl numb of uss hlpin U 1, nd th pmt d is dmpin fcto which cn b st btwn nd 1. W st d to. h. Th dmpin fcto llows th ndom wlk to `scp cycls by jumpin to ndom point in th ntwok th thn followin links fction (1-d) of th tim. ExptisRnk o ER (A) cn b clcultd usin simpl ittiv loithm. Not tht n xptis ntwok could b wihtd. Fo instnc, w cn dd vlus to ds by how fqunt on plis noth. W cn lso wiht ch sk-ply occunc diffntly bsd on how mny plis th in qustion thd. It is stihtfowd to xtnd th notion of Exptis nk to incopot th wihts of th ds by substitutin ER(Ui) with ER(U i )*w ia, wh w ia is th numb of tims i ws hlpd by A nd C(U i )= w ij. In ou pticul study, w found tht wihtin dos not impov th ccucy of ou sults, so fo simplicity w tt th ntwoks s unwihtd, lthouh wihts cn sily b intoducd fo oth pplictions.. HITS Authoity Anoth nkin loithm simil to PRnk is HITS ( Hyptxt inducd topic slction ) [1]. It lso uss n ittiv ppoch, but ssins two scos to ch nod: hub sco nd n uthoity sco. In ou contxt, ood hub is us who is hlpd by mny xpt uss, nd ood uthoity (n xpt) is us who hlps mny ood hubs. Th dfinition is cusiv nd convs ft fw ittions. In ou study, w usd th Authoity vlu of HITS to cospond to th xptis nk of th us.. EVALUATION Sinc th ws no xplicit us-supplid xptis nkin dt in th Jv Foum, w ndd to us humn ts to nt old stndd fo compison. Bcus it ws not possibl fo us to t l numb of ths uss, w ndomly slctd 1 uss fom th ntwok fo us s compison smpl. By omittin thos uss postin fw thn tims, w nsud tht W tid vious vlus (such s. nd.), but it did not mk sinificnt diffnc. th smpld uss hd ntd nouh Foum contnt fo viw to vlut thi xptis lvls. Whil som of th nkin loithms such s ExptisRnk nd HITS cn in pincipl poduc continuous vlus tht cn potntilly diffntit btwn ll uss, it is vy difficult fo humns to sot 1 uss into nkd list. Rts must d fom tn to hundds of msss postd by us to vlut his/h xptis lvl. It is lso difficult to comp two uss whn thy both hv postd mny msss but hv not plid to ch oth. Bsd on ou obsvtion of th foum nd th sults of pilot tin st, w dcidd to ctoiz th uss into xptis lvls instd of complt nkd list. Tbl displys dtils of ths ctoiztions. Tbl : Fiv lvls of xptis tin Lvl Ctoy Dsciption Top Jv xpt Jv pofssionl Knows th co Jv thoy nd ltd dvncd topics dply. Cn nsw ll o most of Jv concpt qustions. Also knows on o som sub topics vy wll, Jv us Knows dvncd Jv concpts. Cn pom ltivly wll. Jv ln Knows bsic concpts nd cn pom, but is not ood t dvncd topics of Jv. 1 Nwbi Just sttin to ln jv. W found two ts who Jv pommin xpts to t th 1 uss' xptis. (Ths xpts w not pt of th sch tm; thy w indpndnt consultnts.).1 Sttisticl Mtics Two of th most fquntly usd coltion msus btwn two nks Spmn s ho nd Kndll s Tu [, 1]. Both of ths mtics hv thi limittions. Th Spmn coltion dos not hndl wk odins wll (wk odin mns tht th multipl itms in th nkin such tht nith itm is pfd ov th oth) nd ou nkins hv lot of wk odins bcus multipl uss ssind th sm tin. Kndll s Tu, on th oth hnd, ivs qul wiht to ny intchn of qul distnc, no mtt wh it occus. Fo instnc, n intchn btwn nk 1 nd will b just s bd s intchn btwn nk nd nk 1. Kndll s Tu my b btt mtic fo ou pupos. Nvthlss, fo th vlution, w psnt both Kndll s Tu nd Spmn s ho. Futhmo, w hv lso ddd TopK mtic, which clcults Kndll s Tu fo only th hihst nks. Aft ch humn t submittd his tins, w tstd th libility of ts by lookin t thi int-t coltion. Th Kndll s Tu distnc btwn th two humn ts ws., nd th Spmn s ho coltion cofficint ws. (p<.1), sufficintly hih t of int-t coltion.. Rsults To hv consvtiv msumnt of th possibl pfomnc fo th utomtic loithms, w futh movd smpls whos tins hv mo thn 1 lvl diffnc btwn th two ts. Th Spmn s ho is. nd Kndll s Tu is.

6 WWW / Tck: E*-Applictions Sssion: E-Communitis btwn th two ts fo th 1 uss lft. (On us ws not td bcus ts potd tht thy didn t hv nouh vidnc.) Thfo, w my xpct tht ny utomtd loithm would t bst chiv ound. coltion with th humn ts. Fo ch of ths uss, in th dt nlysis blow, w summd th tins fom th two ts toth s th stndd humn tin (HR). Fiu shows th sttisticl coltions btwn vious loithms nd th humn tins of th 1 uss. (A snsitivity nlysis includin ll 1 uss showd insinificnt diffncs.) RANK of REPLY N = 1 (). AnswNum 1 1 RANK of INDGR N = (b). Ind RANK of ZTHREADS - RANK of ZDGR - N = N = 11 1 (c). Z_numb (d). Z_d Fiu : Th pfomnc of vious loithms in diffnt sttisticl mtics Fom th fiu, on cn s tht ll of ths nkin loithms iv ltivly hih coltion with th humn-ssind tins. This tlls us tht, indd, stuctul infomtion could b usd to hlp vlut uss xptis in onlin community ntwoks. Supisinly, conty to wht Cmpbll t l. [] nd Dom t l. [] found in thi simultion studis, w found tht, in this l ntwok dt st, ExptisRnk ctully dos not pfom btt thn oth simpl mthods. Instd, th z-sco-bsd nks tnd to poduc slihtly btt sults thn oth mthods. W will tun to this in th subsqunt nlysis, wh w ty to find socil ntwok ftus tht xplin this sult. W cn lso s tht diffnt coltion mtics poduc diffnt sults whn compin th sm dt. Fo instnc, whil Z_d shows th hihst coltion with th TopK mtic, it is th Z_numb tht shows th hihst coltion with th complt Kndll s Tu nd Spmn s Rho mtics. In mny pplictions, w my c mo bout whth th loithm cn idntify th top K xpts, th thn whth it cn t vyon s ltiv xptis. Bin w of ths diffncs in mtics cn hlp on choos n ppopit loithm dpndin on whth it is th top xpts on is ft. W futh lookd t th distibution of utomtic nkins (summizd by th box plots shown in fiu ) cospondin to th humn tin lvls. Fom ths box plots, w cn s th sults consistnt with wht w found in Fiu. W cn s tht th Z_numb, Z_d, nd ExptisRnk ll hv slihtly smll int-qutil n t ch humn tin lvl, which indicts tht thy typiclly hv smll os. W us th tin combintion of two ts h, so th is totl of ctois. RANK of HITS_AUT - N = 11 1 (). HITS_Authoity (f). ExptisRnk Fiu : Box plots of loithm nkins vs. humn tins Whil it is intstin to look t th dtils of ths sults, it is mo impotnt to think bout th bi pictu. W hv obsvd ntwok stuctu diffnt fom th Wb, nd w hv lso sn tht som loithms, such s PRnk nd HITS, which xcl t nkin Wb ps, do not outpfom simpl loithms in this ntwok. Th ky to undstndin th pfomnc of th loithms is in undstndin th humn dynmics tht shp n onlin community. This undstndin will thn hlp slct loithms tht my b mo ppopit fo oth onlin communitis wh th dynmics my b diffnt fom th Jv Foum. Th ppoch w took ws simultion: tkin th simplst st of intction uls tht both plictd th obsvd stuctu nd th ltiv pfomnc of vious loithms. W nxt psnt th sults of thos simultions. N =. SIMULATIONS Much cnt wok on modlin of complx ntwoks in socil, bioloicl nd tchnoloicl domins hs focusd on plictin on o mo t chctistics of l wold ntwoks, such s scl-f d distibutions, clustin, nd v pth lnths[1]. Fo instnc, th pfntil ttchmnt ntwok owth modl of Bbsi t l. [1], wh nw nods joinin pfntilly connct to wll connctd nods, yilds scl-f d distibutions. RANK of PRANK

7 WWW / Tck: E*-Applictions Sssion: E-Communitis H, w tk diffnt ppoch. W plc n mphsis on studyin th vious fctos tht possibly ffct th stuctu of th ntwok. Instd of hvin ttd ntwok to nt, w lt vious fctos dtmin th owth of th ntwok nd obsv how chns in thos fctos ffct th stuctu of th ntwok. Fiu shows snpshot of th simulto w dvlopd to study how ths vious ntwok chctistics (th cospondin contols hiddn in th fiu) will ffct th stuctu of th ntwok in n onlin hlp-skin community nd in tun how thy ffct th pfomnc of vious nkin loithms (shown in th plots nd tbls djcnt to th ntwok lyout). Dtils of this simulto cn b found in []. plyin incss xponntilly with th xptis lvl diffnc btwn th two uss: Exp(L(u) L()) P H (u,) = Exp(L(v) L()) v Not tht ccodin to this fomul, vn us with low lvl of xptis thn th sk hs smll pobbility of nswin th qustion, just s is th cs in th ctul Jv Foum. Aft sttin up th modl, w n th simultion to nt ntwoks. At ch stp, n sk ws pickd to sk qustion nd hlp ws pickd to nsw bsd on th ltd pobbilitis. Aft w n th simultion fo stps, w ot ntwok with th sm v d s th Jv Foum ntwok. Fom scld down vsions, shown in Fiu nd Fiu 1, on cn s tht in this modl, most of links fom low xptis (smll nods in th ntwok visuliztion) to hih xptis (bi nods). Thn, w nlyzd th d distibution of th simultd ntwok to tst whth it ws simil to th Jv Foum ntwok. By compin Fiu with Fiu, on cn s tht whil th ind distibution plicts th hvy skw of th mpiicl ntwok, th outd distibution dos not. Th not s mny sinl-post sks with low outd (, 1, tc) in th simultd ntwok. This is to b xpctd, sinc w not modlin th owth dynmics wh nwcoms, by vitu of not bin in community lon nouh to sk l numb of qustions, contibut to th low nd of th distibution. Whn w updtd ou simultion to llow uss to join th community with som pobbility t in ch stp, w w bl to plict th outd distibution (shown in Fiu ). Fiu : Snpshot of th ntwok simulto intfc.1 Modlin Jv Foum's Ntwok Fom th mpiicl nlysis of th Jv Foum, w incopotd th followin dynmics ovnin th foum into ou modl: Th mjoity of uss md fw posts, ith bcus thy w nw o hd low xptis. Th w numb of xpts who minly nswd oths qustions nd sldom skd qustions thmslvs. Uss smd to nsw oths qustions ccodin to thi own bility cospondin to thi lvl of xptis. Fist, w initilizd th community with 1, uss in th community (on-tnth of th obsvd popultion of th Jv Foum) with pow lw distibution fo th lvls of xptis. Th w mny lvl 1 (novic) uss nd ltivly fw lvl (xpt) uss. Scond, w modld tht low-lvl uss hv hih pobbilitis to sk qustions. A us u with xptis lvl L(u) hs th pobbility to sk qustions P A (u) dtmind by th fomul blow: -1 (L(u) + 1) PA (u) = -1 (L(v) + 1) v Thid, w modld which uss w most likly to nsw qustion posd by us with xptis lvl L() by usin bst pfd xpt ul, wh th pobbility P H (u,) of Fiu : Simultd d distibutions with bst pfd hlps Fiu : Simultd d distibutions with owin ntwok W futh lookd t oth chctistics of th ntwok. Tbl shows tht th bow ti stuctu of th simultd ntwok is simil to th Jv Foum ntwok. Th only sinificnt diffnc is tht w hv ltivly l potion of disconnctd uss. This is bcus in th simultion, w built th ntwok bsd on postin-plyin pttns, but in th Jv Foum, th luks (cospondin to disconnctd nods in ou ntwok) do not post in th community nd thfo not pt of th mpiicl ntwok. Tbl : Bow ti stuctu of th bst pfd ntwok Co In Out Tndils Tubs Disc 1.%.%.%.1% 1.% 1.%

8 WWW / Tck: E*-Applictions Sssion: E-Communitis Fiu 11 shows tht th ind coltion pofil fits th closly with tht of th Jv Foum ntwok. Th coltion btwn sk nd hlp ind is indistinuishbl fom ( =., p =.) o 1 l (. d n i k s 1 hlp ind (loithmiclly binnd) Fiu 11: D coltion pofil of th bst pfd ntwok W tstd vious loithms in this ntwok nd compd thi nks with th nods ssind nk in th simultion pocss. Fiu 1 displys th sult fo popl to mk us of on noth s tim nd xptis []. Such us bhvio ws not modld in ou bst pfd modl. W thus constuctd n ltnt modl, wh uss who hv slihtly btt lvl of xptis thn th sk hv hih pobbility of nswin th qustion, th thn thos with much l diffnc in xptis. This modl uss "just btt" ul, wh us u s pobbility of nswin qustion posd by us is dcidd by th fomul blow: P H (u,) = Exp(L() L(u)) Exp(L() L(v)) v whn L(u)>L() Fiu 1 shows ntwok ntd usin this modl. In contst to th "bst pfd ntwok" shown in Fiu 1, w cn s tht th links not ll pointin to th hihst xpts. Rth, qustions nswd by uss with hih, but not hihst, xptis. Fiu 1: bst pfd ntwok Fiu 1: just btt ntwok Fiu 1: Pfomnc of xptis-dtction loithms on th bst pfd ntwok Fom this fiu, on cn s tht loithms lik ExptisRnk nd HITS do not pfom btt thn simpl mthods lik ind nd z-sco, much lik wht w found mpiiclly in th Jv community. This confims ou intuition tht stuctul diffncs my b mjo son why complx loithms lik ExptisRnk do not lwys wok wll in vious ntwok stuctus.. An Altntiv Ntwok Modl As w sw in th pvious sction, ou simpl modl dynmics cptu both th stuctul ftus nd xptis nkin loithm pfomnc of th ctul Jv Foum. Howv, not ll onlin xptis communitis will follow th sm dynmics s th Jv Foum. W cn ln usful insihts by modlin diffnt dynmics nd thn vlutin th xptis nkin loithms on th modls thy ct. Fo xmpl, in oth communitis, spcilly ons tht my b situtd within n oniztion, xpts my b und tim constints nd choos to nsw only thos qustions tht mk bst us of thi xptis. Thy would thfo b mo likly to nsw th qustions of thos slihtly lss xpt thn thmslvs. It my b th bst wy Fiu 1 shows th d distibution of th ntwok nd Tbl shows th bow ti stuctu nlysis sult. Thy not vy simil to Jv Foum (not th vy tiny Co in th bow ti stuctu), but som pttns clos (such s th hihly skwd d distibution nd th bist bow ti pt bin In ). Fiu 1: Simultd d distibutions with just btt hlps Tbl : bowti nlysis of th just btt ntwok Co 1% In.% Out.% Tndils.% Tubs.% Disc 1.% Fiu 1 shows th d coltion pofil, with n intstin ppnc of ston coltion lon th dionl wh uss hlpin thos slihtly lss xpt thn thmslvs. At.1, th coltion cofficint is positiv in contst to th lck of coltion obsvd in both th mpiicl ntwok nd th bst pfd modl.

9 WWW / Tck: E*-Applictions Sssion: E-Communitis o l ( d n i k s hlp ind (loithmiclly binnd) Fiu 1: Coltion pofil of th just-btt ntwok Fiu displys th pfomnc compisons of th vious nkin loithms in this nw ntwok: ExptisRnk nd Z_sco pfom th bst, nd HITS_uthoity is th wost. Sinc hubs nd uthoitis infoc on noth in th ittiv HITS loithm, in th bst pfd ntwok, th nwbis who hv thi qustions nswd by th bst xpts infoc th scos of thos xpts. Howv, in th just btt loithm, th nwbis who skin th most qustions oftn hlpd by uss with only slihtly hih xptis. Thfo HITS idntifis individuls with mdium xptis s th hihst xpts. Similly Fiu 1 shows n xmpl of hih xpt us who is hlpin oth xpt uss. Sinc xpts hv low HITS hub scos, thy thus impt low HITS uthoity sco to th xpt hlpin thm. On th oth hnd, ExptisRnk popts th xptis sco fom th nwbis to th intmdit uss who nsw thi qustions nd fom th intmdit uss to th bst xpts. Thus w xpct tht PRnk-bsd loithms such s ExptisRnk will in nl outpfom oth loithms whn th sks nd hlps xptis is coltd. Th Jv Foum did not disply this bhvio (in fct, it is ldy vy wll dscibd by ou fist modl). But, s mntiond, such scnio is plusibl wh uss mk th bst us of thi tim by bin mo slctiv in choosin qustions tht chllnin to thm yt thy still cpbl of nswin. Fiu : Pfomnc of xptis nkin loithms in th just btt ntwok 1 Fiu 1: A cs wh hih xptis nod hs low uthoity. SUMMARY AND FUTURE WORK In summy, w wntd to umnt how popl cn hlp on noth in onlin communitis, pticully hlp-skin o tchnicl suppot communitis. To do this, w wishd to umnt wht w cll th xptis ntwok h th wy tht xptis is distibutd nd dployd in pctic. To do this, w wnt thouh th stps. Fist, w wntd to know wht wnt on socilly in typicl hlp-skin community. W nlyzd th ntwok psntin sk-hlp intctions in n onlin community, th Jv Foum. Amon thm w hihly skwd d distibutions, much lik th ph of th Wold Wid Wb. But unlik th Wb, spcific dynmics ovnin this pticul foum poduc diffnt bowti stuctu nd d coltion pofil. W thn n n vlution of xptis nkin loithms loithms to nlyz th ltiv xptis of diffnt uss in this community. To undstnd th sults, w simultd ths dynmics nd poducd ntwoks tht not only mtchd th obsvd t ntwok chctistics but lso llowd us to undstnd why utomtd xptis nkin loithms pfom diffntly in diffntly stuctud ntwoks. This undstndin should hlp us wih th tdoffs in loithm dsin nd us fo ntwoks w ncount in th futu. In fct, it is citicl to do so. In this wok, thn, w found: Stuctul infomtion cn b usd fo vlutin n xptis ntwok in n onlin sttin, nd ltiv xptis cn b utomticlly dtmind usin socil ntwok-bsd loithms. W lso found, howv, tht th ntwok's stuctul chctistics mtt. Ths loithms did nly s wll s humn ts. Howv, th w sinificnt tdoffs mon th loithms. Somtims ltivly simpl msu ws s ood s mo complx loithms, such s n dpttion of PRnk.

10 WWW / Tck: E*-Applictions Sssion: E-Communitis W bliv, nd hv tstd with simultions, tht th stuctul chctistics of th onlin communitis ld to diffncs in th pfomnc of ths loithms. Indictly, w lso dtmind tht simultion is usful mthod fo th nlysis of xptis ntwoks nd xptis findin. W w bl to ti th pfomnc of th loithm dictly bck to th dynmics of th communitis. Th simultions indictd und wht stuctul conditions, o in wht kind of ntwoks, thos loithms will pfom bst. And w w bl to do this without quiin intvntions in l oniztions, xpimntl conditions which w cnnot obtin. Wok mins to b don. Fist, w would lik to look t svl oth hlp-skin communitis (such s n intnt community) nd comp it with ou sults nd simultions. This would nbl us to in mo insihts bout th tdoffs in usin ths loithms s wll s in modlin onlin communitis. Scond, w will xplo loithms tht combin contnt infomtion (to diffntit spcific knowld) nd stuctul infomtion in od to dvlop mo dvncd onlin community bsd xptis finds.. ACKNOWLEDGMENTS This wok hs bn fundd, in pt, by th Ntionl Scinc Foundtion (IRI-). W lso wish to thnk Stv Moow nd Michl Duffy t Jv Foum fo thi contibution, nd Volk Wolf, Jodi Tyon, Go Funs, Michl Cohn, TJ Guili, Y Du, Xiomu Zhou, Kvin Nm, Jin Yoo, Bn Conlton, nd th nonymous viws fo fdbck nd sustions.. REFERENCES 1. Bbsi, A.L. nd Albt, R. Emnc of Sclin in Rndom Ntwoks. Scinc,, -1.. Ackmn, M.S. nd McDonld, D.W. Answ Gdn : min oniztionl mmoy with collbotiv hlp. In Pocdins of CSCW ', Boston, MA,, ACM Pss, -. Ackmn, M.S., Wulf, V. nd Pipk, V. (ds.). Shin Exptis: Byond Knowld Mnmnt. MIT Pss,.. Bkhin, P. A Suvy on PRnk Computin. Intnt Mth. (1),, -1. Bolln, J., d Sompl, H., Smith, J. nd Luc, R. Towd ltntiv mtics of jounl impct: A compison of downlod nd cittion dt. Infomtion Pocssin & Mnmnt, 1 () Boodin, A., Robts, G.O., Rosnthl, J.S. nd Tsps, P. Link Anlysis Rnkin Aloithms Thoy And Expimnts. ACM Tnsctions on Intnt Tchnoloy, (1) Bod, A., Kum, R., Mhoul, F., Rhvn, P., Rjopln, S., Stt, R., Tomkins, A. nd Win, J. Gph stuctu in th Wb. Comput Ntwoks, (1-). -.. Cmpbll, C.S., Mlio, P.P., Cozzi, A. nd Dom, B., Exptis idntifiction usin mil communictions. In th twlfth intntionl confnc on Infomtion nd knowld mnmnt, Nw Olns, LA,, -1. Dom, B., Eion, I., Cozzi, A. nd Zhn, Y., Gph-bsd nkin loithms fo -mil xptis nlysis. In DMKD, Nw Yok, NY,, ACM Pss, -.. Fin, R., Kum, R. nd Sivkum, D., Compin top k lists. In Pocdins of th foutnth nnul ACM-SIAM symposium on Disct loithms, Bltimo, MA,, Socity fo Industil nd Applid Mthmtics, Fish, D., Smith, M. nd Wls, H., You A Who You Tlk To. In HICSS ', Hwii, --.pdf 1. Fon, L.N. Ynt: multi-nt, fl-bsd mtchmkin systm. In Pocdins of Ants ', ACM Pss, Min dl Ry, CA,, 1-1. Hlock, J.L., Konstn, J.A., Tvn, L.G. nd Ridl, J.T. Evlutin collbotiv filtin commnd systms. ACM Tnsctions on Infomtion Systms, (1) Kutz, H., Slmn, B. nd Shh, M. Rfl Wb: combinin socil ntwoks nd collbotiv filtin. Commun. ACM, () Klinb, J.M. Hubs, uthoitis, nd communitis. Acm Computin Suvys, 1. U1-U 1. Kollock, P. Th conomis of onlin cooption: ifts nd public oods in cybspc. In Smith, M.A. nd Kollock, P. ds. Communitis in Cybspc, Routld, London,.. Kulwich, B. nd Buky, C., ContctFind nt: nswin bulltin bod qustions with fls. In th 1th Ntionl Confnc on Atificil Intllinc, Potlnd, OR,, Lkhni, K. nd von Hippl, E. How opn souc softw woks: "f" us-to-us ssistnc. Rsch Policy, (), -. Littlp, G.E. nd Mull, A.L. Rconition nd utiliztion of xptis in poblm-solvin oups: Expt chctistics nd bhvio. Goup Dynmics: Thoy, Rsch, nd Pctic, Liu, X., Bolln, J., Nlson, M.L. nd Sompl, H.V.D. Couthoship ntwoks in th diitl liby sch community. Infomtion Pocssin nd Mnmnt, 1 () Nwmn, M.E.J. Th stuctu nd function of complx ntwoks. Sim Rviw, (). 1-. P, L., Bin, S., Motwni, R. nd Winod., T. Th Pnk Cittion Rnkin: Binin Od to th Wb, Stnfod Diitl Liby Tchnolois Pojct,. Si Mslov, K.S., Alxi Zliznyk. Pttn Dtction in Complx Ntwoks: Coltion Pofil of th Intnt pint Xiv:cond-mt/,. Stt, L. nd Lochbum, K., Who Knows: A Systm Bsd on Automtic Rpsnttion of Smntic Stuctu. In Pocdins of RIAO,, -. Wssmn, S. nd Fust, K. Socil Ntwok Anlysis: Mthods nd Applictions. Cmbid Univsity Pss, Cmbid,. Zhn, J., Ackmn, M.S. nd Admic, L. CommunityNtSimulto: Usin Simultion to Study Onlin Community Ntwok Fomtion nd Implictions, In Pocdins of C&T ', Est Lnsin, MI,

The author(s) shown below used Federal funds provided by the U.S. Department of Justice and prepared the following final report:

The author(s) shown below used Federal funds provided by the U.S. Department of Justice and prepared the following final report: Th author(s) shown blow usd Fdral funds providd by th U.S. Dpartmnt of Justic and prpard th following final rport: Documnt Titl: Author(s): Impact Munitions Data Bas of Us and Effcts Kn Hubbs ; David Klingr

More information

Two Concepts of Causation Ned Hall

Two Concepts of Causation Ned Hall Two Conpts of Custion N Hll 1 Introution Custion, unrstoo s rltion btwn vnts, oms in t lst two bsi n funmntlly iffrnt vritis. On of ths, whih I ll pnn, is simply tht: ountrftul pnn btwn wholly istint vnts.

More information


IMPROVE CUSTOMERS LOYALTY IN ONLINE GAMING: AN EMPIRICAL STUDY Fan Zhao Fan Zhao ABSTRACT In th past dcad, onlin gams hav bcom an important lctronic commrc application A good undrstanding of customr onlin gam bhaviors is critical for both rsarchrs and practitionrs, such as

More information

Quantum Graphs I. Some Basic Structures

Quantum Graphs I. Some Basic Structures Quantum Graphs I. Som Basic Structurs Ptr Kuchmnt Dpartmnt of Mathmatics Txas A& M Univrsity Collg Station, TX, USA 1 Introduction W us th nam quantum graph for a graph considrd as a on-dimnsional singular

More information

Systems of First Order Linear Differential Equations

Systems of First Order Linear Differential Equations Sysms of Firs Ordr Linr Diffrnil Equions W will now urn our nion o solving sysms of simulnous homognous firs ordr linr diffrnil quions Th soluions of such sysms rquir much linr lgbr (Mh Bu sinc i is no

More information

Block. It puts some writers down for DRAFT NO.4. Replacing the words in boxes. THE WNTING UFE

Block. It puts some writers down for DRAFT NO.4. Replacing the words in boxes. THE WNTING UFE Block. It puts som writrs down for months. It puts som writrs down for lif. A not always brif or minor form of it muts all writrs from th outst of vry day. "Dar Jol..." This is just a random sampl from

More information

When Simulation Meets Antichains (on Checking Language Inclusion of NFAs)

When Simulation Meets Antichains (on Checking Language Inclusion of NFAs) When Simultion Meets Antichins (on Checking Lnguge Inclusion of NFAs) Prosh Aziz Abdull 1, Yu-Fng Chen 1, Lukáš Holík 2, Richrd Myr 3, nd Tomáš Vojnr 2 1 Uppsl University 2 Brno University of Technology

More information

Optimization design of structures subjected to transient loads using first and second derivatives of dynamic displacement and stress

Optimization design of structures subjected to transient loads using first and second derivatives of dynamic displacement and stress Shock and Vibration 9 (202) 445 46 445 DOI 0.3233/SAV-202-0685 IOS Prss Optimization dsign of structurs subjctd to transint loads using first and scond drivativs of dynamic displacmnt and strss Qimao Liu

More information

Some Techniques for Proving Correctness of Programs which Alter Data Structures

Some Techniques for Proving Correctness of Programs which Alter Data Structures Some Techniques for Proving Correctness of Progrms which Alter Dt Structures R. M. Burstll Deprtment of Mchine Intelligence University of Edinburgh 1. INTRODUCTION Consider the following sequence of instructions

More information

The Dozenal Society of America

The Dozenal Society of America The Dozenal Society of America Multiplication Tables of Various Bases by Michael Thomas De Vlieger Ever wonder what multiplication tables might look like in alternative bases? The dsa Updated 6 February

More information

On the Robustness of Most Probable Explanations

On the Robustness of Most Probable Explanations On the Robustness of Most Probble Explntions Hei Chn School of Electricl Engineering nd Computer Science Oregon Stte University Corvllis, OR 97330 Adnn Drwiche Computer Science

More information

Magnetism from Conductors, and Enhanced Non-Linear Phenomena

Magnetism from Conductors, and Enhanced Non-Linear Phenomena Mnetism from Conductors, nd Enhnced Non-Liner Phenomen JB Pendry, AJ Holden, DJ Roins, nd WJ Stewrt Astrct - We show tht microstructures uilt from non-mnetic conductin sheets exhiit n effective mnetic

More information

Category 11: Use of Sold Products

Category 11: Use of Sold Products 11 Catgory 11: Us of Sold Products Catgory dscription T his catgory includs missions from th us of goods and srvics sold by th rporting company in th rporting yar. A rporting company s scop 3 missions

More information

Does the chimpanzee have a theory of mind? 30 years later

Does the chimpanzee have a theory of mind? 30 years later Review Does the chimpnzee hve theory of mind? 30 yers lter Josep Cll nd Michel Tomsello Mx Plnck Institute for Evolutionry Anthropology, Deutscher Pltz 6, D-04103 Leipzig, Germny On the 30th nniversry

More information

Financial Ties between DSM-IV Panel Members and the Pharmaceutical Industry

Financial Ties between DSM-IV Panel Members and the Pharmaceutical Industry Regulr Article DOI: 10.1159/000091772 Finncil Ties between DSM-IV Pnel Members nd the Phrmceuticl Industry Lis Cosgrove b Sheldon Krimsky Mnish Vijyrghvn Lis Schneider University of Msschusetts, Boston,

More information

VoIP for the Small Business

VoIP for the Small Business VoIP for the Smll Business Reducing your telecommunictions costs Reserch firm IDC 1 hs estimted tht VoIP system cn reduce telephony-relted expenses by 30%. Voice over Internet Protocol (VoIP) hs become

More information

NAEYC Early Childhood Program Standards and Accreditation Criteria & Guidance for Assessment

NAEYC Early Childhood Program Standards and Accreditation Criteria & Guidance for Assessment NAEYC Erly Childhood Progrm Stndrds nd Accredittion Criteri & Guidnce for Assessment This document incorportes the lnguge of ll NAEYC Erly Childhood Progrm Stndrds nd Accredittion Criteri, including 39

More information

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. Do Artifcts Hve Politics? Author(s): Lngdon Winner Source: Dedlus, Vol. 109, No. 1, Modern Technology: Problem or Opportunity? (Winter, 1980), pp. 121-136 Published by: The MIT Press on behlf Americn Acdemy

More information

What are the most abundant proteins in a cell?

What are the most abundant proteins in a cell? What ar th most abundant s in a cll? Evn aftr rading svral txtbooks on s, on may still b lft wondring which of ths critical molcular playrs in th lif of a cll ar th most quantitativly abundant. Though

More information

All-Atom Empirical Potential for Molecular Modeling and Dynamics Studies of Proteins

All-Atom Empirical Potential for Molecular Modeling and Dynamics Studies of Proteins 3586 J. Phys. Chem. B 1998, 102, 3586-3616 All-Atom Empiricl Potentil for Moleculr Modeling nd Dynmics Studies of Proteins A. D. McKerell, Jr.,*,, D. Bshford,, M. Bellott,, R. L. Dunbrck, Jr.,, J. D. Evnseck,,

More information

TheImpactoftheNation smost WidelyUsedInsecticidesonBirds

TheImpactoftheNation smost WidelyUsedInsecticidesonBirds TheImpctoftheNtion smost WidelyUsedInsecticidesonBirds Neonicotinoid Insecticides nd Birds The Impct of the Ntion s Most Widely Used Insecticides on Birds Americn Bird Conservncy, Mrch 2013 Grsshopper

More information



More information

Accuracy at the Top. Abstract

Accuracy at the Top. Abstract Accuacy at the Top Stephen Boyd Stanfod Univesity Packad 64 Stanfod, CA 94305 Mehya Mohi Couant Institute and Google 5 Mece Steet New Yok, NY 00 Coinna Cotes Google Reseach

More information

On the Algorithmic Implementation of Multiclass Kernel-based Vector Machines

On the Algorithmic Implementation of Multiclass Kernel-based Vector Machines Jounal of Machine Leaning Reseach 2 (2001) 265-292 Submitted 03/01; Published 12/01 On the Algoithmic Implementation of Multiclass Kenel-based Vecto Machines Koby Camme Yoam Singe School of Compute Science

More information


THE INSCRIPTIONS FROM TEMPLE XIX AT PALENQUE DAVID STUART THE INSCRIPTIONS FROM TEMPLE XIX AT PALENQUE The Inscriptions fromtemple XIX t Plenque A Commentry The Inscriptions from TempleXIX t Plenque A Commentry By Dvid Sturt Photogrphs y Jorge Pérez

More information

First variation. (one-variable problem) January 21, 2015

First variation. (one-variable problem) January 21, 2015 First vrition (one-vrible problem) Jnury 21, 2015 Contents 1 Sttionrity of n integrl functionl 2 1.1 Euler eqution (Optimlity conditions)............... 2 1.2 First integrls: Three specil cses.................

More information


LIFE AS POLY-CONTEXTURALITY *) Ferury 2004 LIFE AS POLY-CONTEXTURALITY *) y Gotthrd Günther Kein Leendiges ist ein Eins, Immer ist's ein Vieles. (Goethe) Prt I : The Concept of Contexture A gret epoch of scientific trdition is out to

More information

The 19th Annual International Symposium on Computer Architecture. Alternative Implementations of Two-Level Adaptive Branch Prediction

The 19th Annual International Symposium on Computer Architecture. Alternative Implementations of Two-Level Adaptive Branch Prediction he 19th nnl Intentionl Smposim on Compte hitete pp 124-134, M 19-21, 1992, Gold Cost, stli ltentive Implementtions of wo-level dptive Bnh Pedition se{y Yeh nd Yle Ptt Deptment of Eletil Engineeing nd Compte

More information


CHAPTER 9 THE TWO BODY PROBLEM IN TWO DIMENSIONS 9. Intoduction CHAPTER 9 THE TWO BODY PROBLEM IN TWO DIMENSIONS In this chapte we show how Keple s laws can be deived fom Newton s laws of motion and gavitation, and consevation of angula momentum, and

More information


ARB THERE TImRB COUNTEREXAMPLES TO THE TIm CU)SURE CLOSURE PRINCIPLE? JONATHAN VOGEL ARB THERE TmRB COUNTEREXAMPLES TO THE Tm CU)SURE CLOSURE PRNCPLE? PRNCPLB7 Very ten, person cn't know proposition propositon without wilhout knowing vrious logicl consequences tht ht proposition.

More information